Robust voice activity detection in noisy sound environments is a very difficult problem when using a small device mounted in the ear. Such systems that rely on using fixed thresholds often suffer from false positives and false negatives.